相关论文: The Ghost in the Datacenter: Link Flapping, Topolo…

Node and link churn in multi-party, cross-region clusters over wide-area networks (WANs) often disrupts distributed training. However, checkpoint-based recovery and cloud-centric autoscaling react slowly and assume centralized control,…

Graph Neural Networks (GNNs) have emerged as a powerful machine learning method for graph-structured data. A plethora of hardware accelerators has been introduced to meet the performance demands of GNNs in real-world applications. However,…

Dataset deduplication is widely recognized as a crucial preprocessing step that enhances data quality and improves the performance of large language models. A commonly used method for this process is the MinHash Locality-Sensitive Hashing…

The idle computers on a local area, campus area, or even wide area network represent a significant computational resource---one that is, however, also unreliable, heterogeneous, and opportunistic. This type of resource has been used…

Online LLM inference powers many exciting applications such as intelligent chatbots and autonomous agents. Modern LLM inference engines widely rely on request batching to improve inference throughput, aiming to make it cost-efficient when…

Gated DeltaNet (GDN) is a linear attention mechanism that replaces the growing KV cache with a fixed-size recurrent state. Hybrid LLMs like Qwen3-Next use 75% GDN layers and achieve competitive accuracy to attention-only models. However, at…

Attention, as a core layer of the ubiquitous Transformer architecture, is the bottleneck for large language models and long-context applications. While FlashAttention-3 optimized attention for Hopper GPUs through asynchronous execution and…

In this paper we study the problem of how resilient networks are to node faults. Specifically, we investigate the question of how many faults a network can sustain so that it still contains a large (i.e. linear-sized) connected component…

Tor is a widely used anonymity network that conceals user identities by routing traffic through encrypted relays, yet it remains vulnerable to traffic correlation attacks that deanonymize users by matching patterns in ingress and egress…

When accelerators fail in modern ML datacenters, operators migrate the affected ML training or inference jobs to entirely new racks. This approach, while preserving network performance, is highly inefficient, requiring datacenters to…

Pushback is a mechanism for defending against Distributed Denial-of-Service (DDoS) attacks. DDoS attacks are treated as a congestion-control problem, but because most such congestion is caused by malicious hosts not obeying traditional…

The availability of wide-ranging third-party intellectual property (3PIP) cores enables integrated circuit (IC) designers to focus on designing high-level features in ASICs/SoCs. The massive proliferation of ICs brings with it an increased…

Interdependent networks are characterized by two kinds of interactions: The usual connectivity links within each network and the dependency links coupling nodes of different networks. Due to the latter links such networks are known to…

Modern GPUs adopt chiplet-based designs with multiple private cache hierarchies, but current programming models (CUDA/HIP) expose a flat execution hierarchy that cannot express chiplet-level locality or synchronization. This mismatch leads…

Bots, in recent times, have posed a major threat to enterprise networks. With the distributed nature of the way in which botnets operate, the problems faced by enterprises have become acute. A bot is a program that operates as an agent for…

The outsourcing of the design and manufacturing of integrated circuits (ICs) in the current horizontal semiconductor integration flow has posed various security threats due to the presence of untrusted entities, such as overproduction of…

This paper studies the problem of broadcasting in synchronous point-to-point networks, where one initiator owns a piece of information that has to be transmitted to all other vertices as fast as possible. The model of fractional dynamic…

DNS is important in nearly all interactions on the Internet. All large DNS operators use IP anycast, announcing servers in BGP from multiple physical locations to reduce client latency and provide capacity. However, DNS is easy to spoof:…

Tensor parallelism (TP) enables large language models (LLMs) to scale inference efficiently across multiple GPUs, but its tight coupling makes systems fragile: a single GPU failure can halt execution, trigger costly KVCache recomputation,…

LLM training at the scale of tens of thousands of GPUs now spans multiple datacenters (DC), making cross-DC collectives over long-haul links unavoidable. A critical and overlooked bottleneck arises when these collectives collide with…